Procedures

Recruitment and assignment of interventions

Participants were enrolled through the CLASSIC cohort. In a standard trial, patients receive information and then provide informed consent to participate. At that point, they are randomised. A significant drawback is that patients are told about different treatments in the different arms (including any new treatment), but only half the patients are randomised to that new treatment. This can cause dissatisfaction.

In the cmRCT, patients eligible for the trial are identified from the cohort and randomly selected. Patients who are randomly selected for usual care continue to be followed up in the cohort and are not informed about the trial or the randomisation. Patients who are randomised to the new treatment are then contacted and offered the treatment. They still provide consent to the new treatment and can decide whether or not they wish to receive it. If patients agree to the new treatment, they are provided with the new treatment and continue to be followed up in the cohort. If patients decide that they do not wish to receive the new treatment, they continue to receive usual care and continue to be followed up in the cohort.

We piloted these procedures in 50 patients to test the rate of uptake of the new treatment.

After assessment of eligibility, we selected patients randomly for health coaching or usual care using appropriate central randomisation through a clinical trials unit to ensure concealment of allocation. In this pragmatic evaluation, there was no blinding of patients or providers. All outcomes were either self-reported or routine data.

Sample size and analysis

At the time of study development there were no bespoke methods for powering cmRCTs, and, following existing cmRCTs, we used conventional methods. We powered the study to have 80% power (α=5%) to detect a standardised effect size of 0.25 on any continuous outcome measure. Allowing for 25% attrition among participants–and assuming that outcome measures at baseline correlate 0.5 with their follow-ups –504 patients were needed (252 per arm).

The initial uptake rate was lower than anticipated; hence, we selected a further 252 patients to be offered the intervention. However, within the cmRCT framework all 504 patients offered treatment remained in the treatment group in analysis, including those who declined. In consequence, the effect size between arms detectable at 80% power was 0.39 among those consenting to treatment.

Analysis followed intention-to-treat principles and a prespecified analysis plan (seeAppendix 1). In summary, we report the trial and analysis in accordance with the updated Consolidated Standards of Reporting Trials (CONSORT) standards and utilising the extension for pragmatic trials. The main test of the intervention was that the overall main effect of the intervention is zero. Condition group was used as a binary variable. Continuous outcomes were assessed using linear regression, controlling, where appropriate, for baseline values of the respective outcome. Outcomes measured using ordinal scales were treated as continuous variables. Results for non-normal variables (skew or kurtosis>1.0) were confirmed using bootstrap analysis. Baseline values of outcomes and design factors were included in all analyses. Some additional covariates were prespecified.

Owing to implementation delays, no patient was offered treatment up to 6 months after the baseline assessment and for some the offer was not made until month 12 or later. This caused variations in the duration of time before start of the treatment, ranging from 259 to 513 days. Length of follow-up from

the end of treatment to 20-month follow-up was similarly variable. Thus, the trial is considered to have run for>20 months, with patients receiving treatment at any time within that period.

The cmRCT design provides an estimate of the mean effect in people offered treatment. Compared with a pragmatic trial, which provides an estimate of the mean effect in people agreeing to treatment, the effect is_‘diluted_’by the proportion of patients in the treatment arm who do not consent to treatment. An estimate of the effect size in patients consenting to treatment was obtained through application of a complier-average causal effect (CACE) analysis.105,106_{CACE does not increase the power to detect an effect.}

Economic analysis

The economic analysis aimed to assess the incremental cost-effectiveness of health coaching compared with usual care.

The primary outcome measure for the economic evaluation was health-related quality of life measured by the EQ-5D-5L,107_{a new version developed as a result of concerns over the lack of sensitivity to change} of the original. Published English general population preference weightings70_{were used to convert} responses to a single utility index for each time point.

This was combined with in-hospital mortality information taken from the secondary care utilisation data, applying a utility value of zero to all patients on death. Quality-adjusted life-years (QALYs) were calculated using the‘area under the curve’method, assuming linear extrapolation of utility between time points. QALYs experienced in the second year of the trial were discounted at an annual rate of 3.5%, as specified by the National Institute for Health and Care Excellence (NICE) in its reference case.108

Resource utilisation and costs

Resource utilisation and costs were calculated from the perspective of the UK NHS. Patient-level utilisation data were collected from two sources. Information on GP contacts in the previous 6 months was collected from cohort data at 6, 12 and 18 months. Hospital utilisation data were extracted from linked administrative patient records provided by the NHS, divided into emergency admissions (short stays,≤5 days; long stays, >5 days), elective admissions, elective day cases, outpatients and A&E attendances.

Utilisation data were combined with relevant unit cost data for the price year 2014/15 to calculate total costs. Unit costs not available for this price year were inflated to 2014/15 prices using the Consumer Price Index.109_{Costs occurring in the second year were discounted at a rate of 3.5%.}108

Unit cost figures were sourced from the Personal Social Services Research Unit_’s (PSSRU_’s) unit costs of health and social care (2015)110_{and national NHS reference costs.}111

Health coaching costs

Costs were estimated combining the cost of training and supervising staff, materials and delivery of the health coaching sessions. The intervention was offered to all participants randomly selected, although only 189 received at least one call and were used to estimate costs.

Missing data

Data required for QALY and cost calculation were missing in a small number of cases (n=2), and were imputed. Missing information on age and sex was sourced from administrative data (sex,n=6; age,n=35) or imputed (missing agen=30), to ensure independence from allocation.112

For missing EQ-5D-5L and resource use data, we used multiple imputation by chained equations to generate 50 imputed data sets, assuming that the data were missing at random. The independent variables specified in the imputation models were age, sex, treatment arm and baseline EQ-5D-5L. To account for non-normality, predictive mean matching was used to ensure values observed in the original

package (version 14.2; StataCorp LP, College Station, TX, USA).

Cost–utility analysis

The economic analysis estimates the incremental cost-effectiveness of the offer of health coaching compared with usual care at standard UK willingness-to-pay thresholds.

The primary analysis was based on a comparison on the full sample with multiple imputation. A sensitivity analysis was performed using only the complete-case sample (health coachingn=206, usual caren=378). Analysis used Stata version 14.

The incremental cost-effectiveness ratio (ICER) was calculated, adjusting for age, sex and baseline EQ-5D-5L index score.113_{To assess uncertainty surrounding the estimates and to account for the typically skewed} nature of cost data, incremental costs and QALYs were bootstrapped using pairwise bootstrapping with replacement using 10,000 replications. Cost-effectiveness planes plot these 10,000 bootstrap replications of the ICER estimates to illustrate the uncertainty around the point estimate of the ICER in probabilistic terms. Finally, cost-effectiveness acceptability curves (CEACs) were plotted to represent graphically the probability of the intervention being cost-effective across a range of cost-effectiveness thresholds.

Chapter 9

Results of the CLASSIC cohort

F

Owing to current word limits on the report, we do not present detailed descriptive data on the cohort and restrict the main presentation to quasi-experimental analyses of SICP mechanisms of integration (community assets and care plans).

We present basic descriptive data on patient experience items inAppendix 3. Analyses using the cohort data to explore other aspects of care for older people can be found in published papers,114,115_{and more} will be reported in due course.

A limitation of existing analyses of integrated care is that they are too large in scope or rapid in delivery to allow setting up data collection to capture effects, restricting analyses to routine data that lack patient- reported outcomes. The CLASSIC trial used the cmRCT design to develop a cohort, which provided the ‘context’into which the SICP and its mechanisms of integration would be introduced. The cohort had two functions:

1. to provide a sampling frame for the cmRCT within CLASSIC for formal experimental analyses (full details are provided inChapters 8and13)

2. to provide a sample of the total eligible population of older people, which could be used to track the impact of mechanisms of integration on patients through variation in exposure to those mechanisms among patients in the cohort.

Mailed questionnaires (n = 12,989) Returned questionnaires (n = 4447; 34.2%) Usable questionnaires (n = 4377; 33.6%)

Mailed 6-month follow-up

(n = 4225)

Returned at 12-month follow-up (n = 3390; 77.5%)

Returned at 18-month follow-up (n = 2922; 66.8%) Died (n = 35) Died (n = 26) Excluded as duplicates/not uniquely identifiable (n = 70; 0.5%)

Did not provide address for follow-up (n = 152; 3.5%)

This chapter will focus on the second function. Two mechanisms of integration suited to evaluation through the cohort are community assets and care plans. Community assets were a specific mechanism of integration within the SICP, with its own dedicated workstream (seeChapter 2,Box 1). Care plans have long been seen as critical to effective management of long-term conditions,15,116_{and a major feature of} health policy in the UK.117_{We assessed use of both community assets and care plans in the cohort and} used variation in use to explore their impact on patient outcomes.